U.S. patent application number 14/632009 was filed with the patent office on 2015-09-03 for systems and methods for modeling energy consumption and creating demand response strategies using learning-based approaches.
The applicant listed for this patent is Board of Trustees of The University of Alabama. Invention is credited to Shuhui Li, Ming Sun, Dong Zhang.
Application Number | 20150248118 14/632009 |
Document ID | / |
Family ID | 54006734 |
Filed Date | 2015-09-03 |
United States Patent
Application |
20150248118 |
Kind Code |
A1 |
Li; Shuhui ; et al. |
September 3, 2015 |
SYSTEMS AND METHODS FOR MODELING ENERGY CONSUMPTION AND CREATING
DEMAND RESPONSE STRATEGIES USING LEARNING-BASED APPROACHES
Abstract
According to various implementations, a demand response (DR)
strategy system is described that can effectively model the HVAC
energy consumption of a house using a learning based approach that
is based on actual energy usage data collected over a period of
days. This modeled energy consumption may be used with day-ahead
energy pricing and the weather forecast for the location of the
house to develop a DR strategy that is more effective than prior DR
strategies. In addition, a computational experiment system is
described that generates DR strategies based on various energy
consumption models and simulated energy usage data for the house
and compares the cost effectiveness and energy usage of the
generated DR strategies.
Inventors: |
Li; Shuhui; (Northport,
AL) ; Sun; Ming; (Tuscaloosa, AL) ; Zhang;
Dong; (Tuscaloosa, AL) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Board of Trustees of The University of Alabama |
Tuscaloosa |
AL |
US |
|
|
Family ID: |
54006734 |
Appl. No.: |
14/632009 |
Filed: |
February 26, 2015 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
61944669 |
Feb 26, 2014 |
|
|
|
Current U.S.
Class: |
700/295 ; 706/12;
706/25 |
Current CPC
Class: |
G06N 3/006 20130101;
F24F 11/30 20180101; F24F 11/62 20180101; G05B 13/04 20130101; F24F
11/46 20180101; G06N 3/084 20130101 |
International
Class: |
G05B 13/04 20060101
G05B013/04; F24F 11/00 20060101 F24F011/00; G05F 1/66 20060101
G05F001/66; G06N 99/00 20060101 G06N099/00; G06N 3/08 20060101
G06N003/08 |
Goverment Interests
GOVERNMENT LICENSE RIGHTS
[0002] This invention was made with government support under NSF
#1059265 project awarded by U.S. National Science Foundation. The
government has certain rights in the invention.
Claims
1. An energy consumption management system comprising: one or more
local receivers disposed adjacent a respective one of one or more
energy consuming units in a building, each local receiver
comprising a processor configured for receiving usage instructions
for the adjacent energy consuming unit and causing the usage
instructions to be executed for the energy consuming unit; and a
central computing system comprising: a memory configured for
storing actual usage data associated with at least one energy
consuming unit in the building for one or more time windows on each
of days i through j, wherein day i is the first day for which data
is stored and day j is the most recent day for which data is
stored; and a processor configured for: receiving the actual usage
data from the memory; executing at least one computer-based
learning system to model energy consumption for day j+1 based on at
least the actual energy usage data for the energy consumption unit;
generating a demand response strategy for the energy consuming unit
for day j+1 based on the modeled energy consumption and next-day
energy pricing for each time window for day j+1; and communicating
the demand response strategy of the energy consuming unit to the
local receiver associated with the energy consuming unit, the
demand response strategy comprising the usage instructions.
2. The system of claim 1, wherein the computer-based learning
system is a neural network system.
3. The system of claim 1, wherein the computer-based learning
system is a regression based system.
4. The system of claim 1, wherein causing the usage instructions to
be executed for the energy consuming unit comprises communicating
the usage instructions to the energy consuming unit.
5. The system of claim 1, wherein causing the usage instructions to
be executed for the energy consuming unit comprises controlling the
energy consuming unit according to the usage instructions.
6. The system of claim 1, wherein the demand response strategy
comprises scheduled usage for the energy consuming unit during each
time window of day j+1.
7. The system of claim 1, wherein the energy consuming unit is a
HVAC system and the demand response strategy comprises thermostat
settings for the HVAC system for each time window of day j+1.
8. A system for creating a demand response strategy for a building,
the system comprising a computing device comprising: a memory
configured for storing actual usage data associated with an energy
consuming unit in the building for one or more time windows on each
of days i through j, wherein day i is the first day for which data
is stored and j is the most recent day for which data is stored;
and a processor configured for: receiving the actual usage data
from the memory; executing at least one computer-based learning
system to model energy consumption for day j+1 based on at least
the actual usage data for the energy consumption unit, and
generating a demand response strategy for the energy consuming unit
for day j+1 based on the modeled energy consumption and next-day
energy pricing for each time window for day j+1.
9. The system of claim 8, wherein: the energy consuming unit is a
regulatable energy consuming unit; and the processor is further
configured for: receiving usage data associated with operating a
deferrable load energy consuming unit for each run cycle, receiving
next-day energy pricing for each time window for day j+1, and
generating a demand response strategy for the deferrable load
energy consuming unit, the demand response strategy for the
deferrable load energy consuming unit comprising at least one time
window in day j+1 during which operation of the deferrable load
energy consuming unit is allowed based at least on the usage data
associated with the deferrable load energy consuming unit and the
next-day energy pricing for day j+1.
10. The system of claim 9, wherein the processor is further
configured for electrically communicating the demand response
strategy for day j+1 associated with the regulatable energy
consuming unit to a first control receiver and the demand response
strategy for day j+1 associated with the deferrable load energy
consuming unit to a second control receiver, the first control
receiver being associated with and in electrical communication with
the regulatable energy consuming unit and the second control
receiver being associated with and in electrical communication with
the deferrable load energy consuming unit, the first and second
control receivers configured for controlling operation of the
regulatable energy consuming unit and the deferrable load energy
consuming unit, respectively, based on the communicated demand
response strategies.
11. The system of claim 9, wherein generating the demand response
strategy for the deferrable load energy consuming unit comprises
executing a binary integer programming strategy.
12. The system of claim 8, wherein: the memory is configured for
storing a current charge for a battery on day j, the battery
configured for receiving and storing energy generated by a
photovoltaic system; and the processor is further configured for:
receiving the current charge for the battery from the memory, and
generating a control and management strategy for the battery for
day j+1 based at least on the current charge and next-day energy
generation and consumption difference between at least one
renewable energy source and at least one energy consuming unit,
wherein the control strategy for the battery comprises identifying
one or more time windows during which the battery is charged when
PV generation is higher than the demand or energy from the battery
is discharged for consumption by one or more energy consuming units
when PV generation is below a predetermined level.
13. The system of claim 8, wherein: the energy consumption unit is
an HVAC system, the actual usage data comprises an outdoor
temperature for each time window on days I through j, an indoor
temperature for each time window on days i through j, and a
thermostat setting for each time window on days i through j, and
the processor is further configured for using one or more of the
computer based learning systems to model energy consumption of the
HVAC system for day j+1 based on the weather forecast for day j+1
and the actual usage data for days i through j.
14. The system of claim 13, wherein generating the demand response
strategy comprises generating thermostat settings for one or more
time windows on day j+1 such that energy costs for day j+1 are
minimized.
15. The system of claim 14, wherein the computer-based learning
system comprises a neural network system.
16. The system of claim 15, wherein the neural network system
comprises a multilayer perceptron, and the multilayer perceptron
comprises: first, second, and third input nodes, the first input
node configured for receiving from the memory the outdoor
temperature for each time window during each day i through j, the
second input node configured for receiving from the memory the
indoor temperature for each time window for each day i through j,
and the third input node configured for receiving from the memory
the thermostat setting for each time window for each day i through
j; a computation layer comprising a plurality of computation nodes,
the processor configured for propagating signals from the input
nodes through the computation layer in a forward direction on a
layer-by-layer basis; and an output layer comprising at least one
output node, the output node indicating energy consumption for the
HVAC system for each time window for day j+1.
17. The system of claim 14, wherein the computer-based learning
system comprises a regression based system.
18. The system of claim 17, wherein the regression based system
solves the following non-linear equation to learn energy
consumption of the HVAC system:
Q.sub.k=q(T.sub.k+1.sup.I,T.sub.k.sup.I,T.sub.k.sup.O,.beta.)
wherein Q.sub.k is the k.sup.th HVAC energy consumption observation
stored in the memory, q(.cndot.) is a third order polynomial
function of a predictor variable vector
(T.sub.k+1.sup.I,T.sub.k.sup.I,T.sub.k.sup.O) which includes
thermostat settings, indoor temperatures, and outdoor temperatures
for observations k=1 . . . n, respectively, n is the number of
observations saved in the memory, and .beta. is a parameter vector
that includes .beta..sub.0, .beta..sub.1, . . . .beta..sub.p.
Description
CROSS REFERENCE TO RELATED APPLICATION
[0001] This application claims priority to U.S. Provisional Patent
Application No. 61/944,669 filed Feb. 26, 2014, and entitled
"Systems and Methods for Modeling Energy Consumption and Creating
Demand Response Strategies Using Learning-Based Approaches," the
content of which is herein incorporated by reference in its
entirety.
BACKGROUND
[0003] Electricity consumption in residential markets will undergo
fundamental changes in the next decade due to the emergence of
smart appliances and home automation. A key requirement for the
smart appliances within the smart grid framework is the demand
response (DR). The North American Electric Reliability Corporation
has defined demand response as changes in electricity usage by
end-use customers from their normal consumption patterns in
response to changes in the price of electricity (price-responsive
DR), or to incentive payments designed to induce lower electricity
use at time of high wholesale market prices or when system
reliability is jeopardized (curtailable DR).
[0004] Electric utility companies typically use hourly real-time
price (RTP) or day-ahead price (DAP) structure in their dynamic
pricing programs. In North America, Ameren Focused Energy, serving
about 2.4 million electric customers in Illinois and Missouri, has
very detailed RTP and DAP tariffs posted on their website since
Jun. 1, 2008 for both day-ahead and real-time markets. The
day-ahead market produces financially binding schedules for the
production and consumption of electricity one day before the
operating day. The real-time market reconciles any differences
between the amounts of energy scheduled day-ahead and the real-time
load, market participant re-offers, hourly self-schedules,
self-curtailments and any changes in general, real-time system
conditions. Therefore, the DAP structure provides valuable
information for price-sensitive loads while the RTP structure gives
useful information for curtail operations.
[0005] For a typical home in the United States, home appliances are
responsible for an important part of the energy bills. These
appliances may include home heating, ventilation, and air
conditioning system (HVAC), water heaters, clothes washers and
dryers, dishwashers, refrigerator and freezers, electric stoves
and/or ovens, coffee maker, home electric drive vehicle charging
system, and lights, for example. For each energy consuming
appliance, key factors affecting household energy consumption
include: 1) appliance load level, 2) when and how long an appliance
is used, and 3) how much unwanted heat could be generated when
using the appliance. For a flat electricity price structure, a
customer would use an appliance whenever it is needed. However, for
a dynamic electricity price structure, customers are encouraged to
optimize energy consumption of their DR capable appliances.
[0006] Typically, the HVAC system is one of the more challenging
appliances for a DR strategy. Traditionally, the thermostat of a
HVAC unit is set at 71.degree. or 72.degree. for a typical house in
the United States. However, in a dynamic price framework, the
thermostat setting is regulated according to the real-time price
information.
[0007] In modeling energy consumption of a residential house, the
amount of energy consumed by the HVAC system is typically the most
dominant part. The heat load that the HVAC system must overcome is
mainly generated in three ways: conduction, convection, and
radiation. In most conventional DR studies, the heat load of a
residential house is computed based on simplified approaches that
typically only consider conduction. However, actual energy
consumption of a residential house is much more complicated, which
can be affected by geographical location, design architecture,
window arrangements, insulation materials, occupants, weather,
season, etc. and can change from one day to another.
[0008] A key component for a successful price-responsive DR program
is a home automation system (HAS). Basically, a HAS receives
information about weather forecast, dynamic electricity pricing,
device operating characteristics, usage requests, etc., and
autonomously makes control decisions and sends control actions to
smart appliances.
[0009] However, a great challenge for the HAS to establish an
optimal price-responsive DR strategy is how to accurately model and
estimate the energy consumption of a residential house in variable
weather conditions and a dynamic pricing environment. As noted
above, in existing technologies, many DR techniques are developed
based on simplified energy consumption models. For example, a
strategy to minimize the cost for electricity consumption has been
proposed in which the energy consumption of a house is modeled
based on simple conduction heat transfer equations. For example, a
simplified equivalent and thermal parameter (ETP) modeling approach
is used in GridLAB-D, a distribution system simulator, to estimate
thermal loads of a residential house based on first principles.
Further, a quasi-steady-state approach has been adopted to estimate
hourly building electricity demand, in which the building thermal
model is built based on an equivalent resistance-capacitance
network. The hourly energy consumption is determined through an
optimization strategy under the constraints of several predefined
customer load levels which include maximum and minimum hourly
demands, minimum daily consumption, and ramping up and down
limits.
[0010] Other approaches to DR strategies include a heuristic
approach and an "optimal" DR approach. FIG. 1 illustrates a flow
diagram of a heuristic DR strategy for operation of an HVAC system
during the summer. The heuristic DR strategy is a variable
temperature setting approach. During the summer time, for instance,
the air conditioner should be operated the coolest possible near
the lower boundary, T.sub.min of the ASHRAE summer comfort zone
when the RTP is lower than a predefined value. On the other hand,
the HVAC is operated the hottest possible near the upper boundary,
T.sub.max, of the ASHRAE summer comfort zone. The HVAC is operated
between these two boundaries depending on the RTP tariff. In FIG.
1, P.sub.real is the RTP for the current time frame i. P.sub.min is
a minimum price point, P.sub.max is a maximum price point, and
P.sub.1-P.sub.n are intermediate price points between P.sub.min and
P.sub.max.
[0011] Assuming there are n temperature settings between T.sub.max
and T.sub.min, then, the price and thermostat settings for summer
time are calculated by equations (1) and (2) below.
P.sub.i=P.sub.max-PR.sub.diff tan h(kiPR.sub.diff) (1)
T.sub.i=T.sub.max-(T.sub.max-T.sub.min)/ni (2)
T.sub.i=T.sub.max+(T.sub.max-T.sub.min)/ni (3)
The basic concept is that the thermostat setting is determined
through a combined consideration of maximum and minimum real-time
price and price distribution over a day. In equation (1) above, k
is a constant obtained from a price distribution study for
different seasons, and PR.sub.diff=P.sub.max-P.sub.min, where
P.sub.max and P.sub.min correspond to maximum and minimum
electricity prices of a day, respectively. For winter time, a
modification is necessary with T.sub.max and T.sub.min
corresponding to P.sub.min and P.sub.max, respectively, and the
price and thermostat settings are calculated by equations (1) and
(3) above. Similarly, T.sub.1, T.sub.2, and T.sub.n correspond to
P.sub.1, P.sub.2, and P.sub.n, respectively.
[0012] The DR strategy obtained according to FIG. 1 may be good
some days but bad other days. Another problem with the heuristic DR
strategy is that the algorithm is unable to take the advantage of
low price periods to pre-cool down a house significantly.
[0013] The optimal DR approach attempts to determine the optimal
thermostat setting strategy for a given dynamic price of a day, and
thus is more efficient than the heuristic DR approach. To develop
an optimal DR strategy, the following nonlinear programming
formulation is used:
Minimize : C = i p i Q i Subject to : 0 .ltoreq. Q i .ltoreq. Q max
, T min .ltoreq. T i I .ltoreq. T max ( 4 ) ##EQU00001##
where T.sub.min=T.sub.ideal-d, T.sub.max=T.sup.deal+d, d is the
acceptable temperature deviation, i represents a time slot in one
hour, T.sub.i.sup.I is the room temperature in hour i, C is the
electricity cost during a day, p.sub.i stands for the electricity
price in hour i, Q.sub.i signifies the energy consumed by the HVAC
unit in hour i, and Q.sub.max denotes the maximum energy that can
be consumed by the HVAC unit. Traditionally, the energy consumed by
the HVAC Q.sub.i is modeled as a function of room temperature
T.sub.i.sup.I and outdoor temperature T.sub.i.sup.O. Equation (5)
below shows a simplified thermal model of a residential house:
T.sub.i+1.sup.I=.epsilon.T.sub.i.sup.I+(1-.epsilon.)(T.sub.i.sup.O-.eta.-
Q.sub.i/A) (5)
where .eta. is the efficiency of the HVAC unit, .SIGMA. is the
system inertia, and A is the thermal conductivity. However, in
reality, the relationship between indoor and outdoor temperatures
and energy consumed by the HVAC unit is much more complicated than
Equation (5) so that actual energy consumption of the HVAC unit
could deviate greatly from results generated by using Equation (5),
which affects the DR efficiency. Therefore, an intelligent
mechanism that can identify and update an energy consumption model
daily for a residential house is needed for developing an optimal
DR strategy.
BRIEF SUMMARY
[0014] According to various implementations, a DR strategy
development system is described that can effectively model the
energy consumption of regulatable energy consumption units in a
house using a learning based approach that is based on actual
energy usage data collected over a period of days. This modeled
energy consumption may be used with day-ahead energy pricing and
the weather forecast for the location of the house to develop a DR
strategy that is more effective than prior DR strategies.
[0015] According to certain implementations, a system for creating
a demand response strategy for a building is provided. The system
includes a computing device that includes a memory and a processor.
The memory is configured for storing actual usage data associated
with an energy consuming unit in the building for one or more time
windows on each of days i through j, wherein day i is the first day
for which data is stored and j is the most recent day for which
data is stored. The processor is configured for: (1) receiving the
actual usage data from the memory; (2) executing at least one
computer-based learning system to model energy consumption for day
j+1 based on at least the actual usage data for the energy
consumption unit, and (3) generating a demand response strategy for
the energy consuming unit for day j+1 based on the modeled energy
consumption and next-day energy pricing for each time window for
day j+1. The computer-based learning system may include a neural
network system or a regression based system, for example.
[0016] In one implementation, the energy consumption unit is an
HVAC system, and the actual usage data includes an outdoor
temperature for each time window on days i through j, an indoor
temperature for each time window on days i through j, and a
thermostat setting for each time window on days i through j. In
this implementation, the processor is further configured for using
one or more of the computer based learning systems to model energy
consumption of the HVAC system for day j+1 based on the weather
forecast for day j+1 and the actual usage data for days i through
j. In addition, generating the demand response strategy includes
generating thermostat settings for one or more time windows on day
j+1 such that energy costs for day j+1 are minimized.
[0017] According to another implementation, an energy consumption
management system is provided. The energy consumption management
system includes one or more local receivers and a central computer
computing system. The one or more local receives are disposed
adjacent a respective one of one or more energy consuming units in
a building, and each local receiver includes a processor configured
for receiving usage instructions for the adjacent energy consuming
unit and causing the usage instructions to be executed for the
energy consuming unit. The central computing system includes a
memory and a processor. The memory is configured for storing actual
usage data associated with at least one energy consuming unit in
the building for one or more time windows on each of days i through
j, wherein day i is the first day for which data is stored and day
j is the most recent day for which data is stored. The processor is
configured for: (1) receiving the actual usage data from the
memory; (2) executing at least one computer-based learning system
to model energy consumption for day j+1 based on at least the
actual energy usage data for the energy consumption unit; (3)
generating a demand response strategy for the energy consuming unit
for day j+1 based on the modeled energy consumption and next-day
energy pricing for each time window for day j+1; and (4)
communicating the demand response strategy of the energy consuming
unit to the local receiver associated with the energy consuming
unit. The demand response strategy comprises the usage
instructions. The computer-based learning system may include a
neural network system or a regression based system, for
example.
[0018] According to yet another implementation, a computational
experiment system for comparing demand response strategies is
provided. The system includes a computing device, and the computing
device includes a memory and a processor. The memory is configured
for storing simulated usage data associated with an energy
consuming unit in a building for one or more time windows on each
of days i through j, wherein day i is the first day for which data
is stored and j is the most recent day for which data is stored.
The processor is configured for: (1) receiving the simulated usage
data from the memory; (2) executing at least one computer-based
learning system to model energy consumption for day j+1 based on
the simulated usage data; (3) generating a first demand response
strategy for day j+1 based on the modeled energy consumption and
next day energy price data; (4) generating a second demand response
for day j+1; and (5) generating a display indicating energy costs
associated with the first and second demand response strategies for
day j+1. The computer-based learning system may include a neural
network system, a regression based system, a non linear programming
approach, and/or a particle swarm optimization approach, for
example.
[0019] The second demand response strategy may be generated using a
heuristic approach based on real time pricing data, according to
some implementations. In other implementations, the energy
consuming unit is an HVAC unit having an efficiency .eta., system
inertia .epsilon., and thermal conductivity A, and the second
demand response strategy is based on the following simplified
thermal model approach:
T.sub.i+1.sup.I=.epsilon.T.sub.i.sup.I+(1-.epsilon.)(T.sub.i.sup.O-.eta.-
Q.sub.i/A).
In other implementations, the second demand response strategy may
be generated using a nonlinear programming approach or a particle
swarm optimization approach.
[0020] The processor may also be configured for: (1) receiving an
energy price for each time window on day j+1, (2) generating a
demand response strategy for the energy consuming unit for day j+1
based on the energy price for each time window on day j+1, (3)
calculating a respective cost for operating the energy consuming
unit on day j+1 based on the first and second demand response
strategies and the energy pricing, and (3) displaying a projected
cost for energy usage for the energy consumption unit associated
with each respective schedule associated with each energy
consumption model.
BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The components in the drawings are not necessarily to scale
relative to each other and like reference numerals designate
corresponding parts throughout the several views:
[0022] FIG. 1 illustrates a flow chart of a heuristic DR strategy
for operation of an HVAC system during the summer.
[0023] FIG. 2 illustrates a schematic diagram of a DR strategy
system for an HVAC unit according to one implementation.
[0024] FIG. 3 illustrates a schematic network diagram of the DR
strategy system of FIG. 2.
[0025] FIG. 4 illustrates a schematic diagram of an exemplary
central server configured to implement at least a portion of the DR
strategy system described in FIGS. 2 and 3, according to one
implementation.
[0026] FIGS. 5A-5B illustrate a flow diagram of steps taken by
various portions of the DR strategy system according to one
implementation.
[0027] FIG. 6 illustrates a schematic diagram of the flow of data
through a neural network according to one implementation.
[0028] FIG. 7 illustrates a schematic diagram of an exemplary
central server configured to implement at least a portion of a
computational experiment system according to one
implementation.
[0029] FIGS. 8A-8B illustrate a flow diagram of steps taken by
various portions of the computational experiment system according
to one implementation.
[0030] FIG. 9 illustrates a schematic diagram of a computational
experiment system according to one implementation.
[0031] FIGS. 10-15 illustrate charts comparing simulated and
modeled energy usage and energy pricing using various approaches
according to various implementations.
[0032] FIG. 16 is a schematic diagram illustrating the flow of
energy and information among various energy consuming and
generating units and a home energy management system according to
one implementation.
[0033] FIG. 17 is a schematic diagram illustrating several
components of a solar energy system for a building, according to
one implementation.
[0034] FIG. 18 is a schematic diagram illustrating several
components of the home energy management system according to one
implementation.
[0035] FIG. 19 illustrates a room inside of a building having
various electrical components and sensors, according to one
implementation.
DETAILED DESCRIPTION
[0036] According to various implementations, a DR strategy system
is described that can effectively model the energy consumption and
generation of a house using a learning based approach that is based
at least in part on actual energy usage data collected over a
period of days. This modeled energy consumption may be used with
day-ahead energy pricing and a weather forecast for the location of
the house to develop a DR strategy that is more effective than
prior DR strategies. One exemplary goal of the DR strategy system
is to minimize net energy cost to the consumer. This goal may be
met by reducing the amount of energy consumed during peak tariff
time periods and identifying optimal time periods for storing,
using, or selling energy generated from renewable energy sources
associated with the house.
[0037] In particular, interactive learning mechanisms are disclosed
that are configured to learn and model energy consumption of a
residential house using substantially real-time measured or
simulated energy consumption data. Due to the learning, the model
can accurately capture the thermal behavior of a house for
different seasons, users, and weather conditions. For example, the
thermal storage characteristics of the house may be captured by the
learning mechanisms. The model may then be used to develop a cost
effective DR strategy for one or more energy consuming units in the
house, such as the HVAC system, other appliances, or other energy
consuming devices in the house, and any energy generating units
associated with the house.
[0038] To more accurately model the energy consumption for a
household and identify an optimal DR strategy for the energy units
in the home, energy usage and generation from regulatable loads,
fixed deferrable loads, and future fixed non-deferrable loads and
sources may be optimized separately and then combined to achieve a
final, robust optimal solution. Regulatable loads includes energy
loads for which energy usage may be regulated but not delayed.
Examples of regulatable loads include the HVAC system, hot water
heater(s), and battery(s) storing electricity generated from
renewable energy sources associated with the house (e.g., solar
photovoltaic panels, such as roof-top solar photovoltaic panels,
and wind turbines). Fixed deferrable loads includes energy loads
for which energy usage may be deferred but not regulated. Examples
of fixed deferrable loads include the clothes washer and dryer,
dishwasher, and vehicle charging station. Future fixed
non-deferrable loads include energy loads for which energy usage or
generation cannot be regulated or deferred. Examples of future
fixed non-deferrable loads include lights, microwave, home office
equipment, refrigerator, oven/stove, and energy generated from any
renewable energy sources associated with the house.
[0039] Energy units having regulatable loads and fixed deferrable
loads are subject to management by the HAS using a DR policy
generated the day before. The DR policy may be adjusted in the
current day based on real time pricing (RTP) data, if available.
For example, the DR policy for a regulatable load may be solved
using linear or nonlinear programming methods, and DR policy for
fixed deferrable loads may be solved using an integer programming
technique (e.g., binary integer programming).
[0040] However, energy usage by energy units having future fixed
non-deferrable loads may not be deferred or regulated. Instead,
energy usage by at least a portion of these future fixed
non-deferrable load energy units may be optimized using sensor
technology as part of the DR policy for minimizing energy usage.
For example, a sensor may be associated with one or more energy
consuming units having a future fixed non-deferrable load that
turns off the unit when it is not needed or after a pre-defined
time period or switches the energy consuming unit into an energy
saving or sleep mode when the HEMS 10 senses that no people are in
the room or senses that the appliance is not in use. The
implementation shown in FIG. 19 includes a motion sensor, a door
control sensor, and a security sensor, for example.
[0041] The amount of energy generated by a photovoltaic (PV) system
or other renewable resource (e.g., wind turbines) may be
communicated to the HAS 14, and the HAS 14 may determine whether to
use, store, or sell the energy generated based on the DR policy for
the regulatable and deferrable energy loads in the home, the RTP
tariff for selling energy to the electrical grid, and the remaining
capacity of the battery used to store the energy for later use. In
some implementations, the HAS 14 may treat the energy generated by
the renewable resource as a future fixed generation source with a
HAS generated dynamic pricing and the energy stored by the battery
7 as a regulatable load that is negative when discharging and
positive when charging with a HAS generated dynamic pricing.
[0042] The HAS 14 may also include strategies to learn energy
generation properties of renewable energy source systems, such as
PVs and wind turbines, energy storage characteristics of batteries,
and dynamic price tariffs for the home energy generation system and
battery storage system so that the energy generation system and
energy storage can be operated and coordinated with usage of other
appliances for the maximum benefit of the home owner and to achieve
the goal of having a zero energy home capability as much as
possible.
[0043] For example, various implementations provide an optimal DR
strategy for a household heat pump, or HVAC unit, that is based on
day-ahead weather prediction, day-ahead electricity pricing, and a
learned energy consumption model of the house. The DR strategy for
the HVAC unit for the next day includes the thermostat settings for
various time periods for the next day. The DR strategy for other
appliances or energy consuming units having a fixed deferrable load
may include a RTP tariff range for which operation of the unit is
acceptable, a certain time period during which the appliance or
energy consuming unit may be operated, and/or an acceptable
operation cycle.
[0044] FIGS. 2, 3, and 16-18 illustrate various components of a
home energy management system (HEMS) 10 according to various
implementations. In particular, the system 10 includes a wireless
home-area network 12 with an intelligent HAS 14 and one or more
distributed wireless receivers 16a, 16b, 16c associated with each
energy consuming unit 18a, 18b, 18c, such as an appliance, electric
drive vehicle (EV) charging system, one or more lights, or home
office equipment. The HEMS 10 may also include one or more wireless
sensors 19 associated with energy units having future fixed
non-deferrable loads. The HEMS 10 manages energy usage for the
energy consuming and generating units in the house.
[0045] The HAS 14 receives information from each distributed, local
receiver 16a-16c about energy usage of the energy consuming unit
18a-18c, respectively, over the home area network 12 and stores the
data in a memory, such as queues 26 or other data structures, on a
first in, first out (FIFO) basis. The queues 26 are configured to
hold data for days 1 through n. In one implementation, n may be
eight days. The data is collected by the HAS 14 in substantially
real-time, such as every hour, minute, second, or other appropriate
time period. The HAS 14 may also receive inside and outside
temperatures 25 and thermostat settings for various time periods
during the current day. The HAS 14 also receives predicted weather
20 and electricity price information 22 for the next day via an
external network 24, such as the Internet.
[0046] To implement the DR policy for fixed deferrable loads, the
HAS 14 may include a control block and a next-day DR block. The
control block for the fixed deferrable loads provides control
commands to local receivers at the present day while a local
receiver defers the operation of an energy consuming unit, based on
the control commands received from the HAS 14. The next-day DR
block for fixed deferrable loads may include an optimization module
that receives DAP tariff information from the utility company over
a network, such as the Internet or private network, or a smart
meter attached to the house as well as day-ahead dynamic price
tariffs determined by the HAS 14 for home solar PV system and
battery storage system and executes an optimization routine to
identify the best time slot to operate a deferrable unit for the
next day. In certain implementations, an integer programming
strategy is used to identify the most cost effective time slot(s)
to operate the fixed deferrable load units. Electric utility DAP
tariff information may include other non-RTP tariffs, such as time
of use (ToU) tariffs and critical peaking price (CPP) tariffs.
[0047] The HAS 14 may include a control block and a next-day DR
block for regulatable loads. The control block for regulatable
loads provides control commands to local receivers at the present
day based on a DR policy generated one day ahead while a local
receiver regulates the operation of a regulatable energy unit,
depending on the control commands received from the control block.
The next-day DR block for regulatable loads receives information
from a local receiver about energy usage of an appliance at the
present day as well as predicted weather and electricity price
information and dynamic price tariffs determined by the HAS for the
home solar PV system and battery storage system one day ahead and
determines a DR policy for the next day. The DR policy determined
by the next-day DR block for regulatable loads implements the
demand response of those units for the maximum benefit of the
residential consumer at any weather condition. The ability to
identify the energy consumption models of those units through
data-driven learning mechanisms, such as those described above and
shown in FIG. 2, is an important step for the next-day DR block,
according to certain implementations. For example, an intelligent
learning module of the HAS 14 models energy consumption for one or
more regulatable loads based on the data stored in the queues
26.
[0048] The next-day DR block for regulatable loads may include an
optimization module that creates a DR strategy for the next day
that is based on the modeled next-day energy consumption and
day-ahead energy pricing. For example, the DR strategies for the
next day for regulatable units, such as the HVAC unit and the hot
water heater, include a plurality of operational settings (e.g.,
thermostat settings for the HVAC unit and/or hot water heater)
corresponding to various time periods for the next day. The
next-day DR block for fixed deferrable loads may also include an
optimization module that creates a DR strategy using the DAP or RTP
data and time windows and price ranges that are predefined by the
customer as being acceptable for operation of one or more of the
other energy consuming units. The queues 26 may also store data
from these energy consuming units. The optimization modules may be
further configured for adjusting the DR strategy for the current
day using real-time energy pricing, if available.
[0049] Control commands based on the DR strategy are communicated
by the HAS 14 to local receivers 16a-16c via the home area network
12. The local receivers 16a-16c may discontinue the operation of
their associated energy consuming unit 18a-18c to a later time
period having a cheaper energy price or regulate the operation of
the associated energy consuming units 18a-18c, depending on the
control commands from the HAS 14. FIG. 18 illustrates an exemplary
implementation in which the local receivers 16a-16c include ZIGBEE
nodes and smart switches. However, in other implementations, other
types of receivers may be used.
[0050] The substantially real-time DR system 10 is adaptive to
implement the optimal demand response strategy for the maximum
benefit of a residential consumer insubstantially any weather
conditions. A key component of various implementations of the
system is the ability to identify a substantially real-time HVAC
energy consumption model that is adaptive to weather and seasonal
conditions through learning-based computing approaches.
[0051] The HEMS 10 may also be configured for reporting usage
conditions to a user, allow the user to activate or deactivate DR
appliances, adjust temperatures (or ranges thereof) for regulatable
appliances, accept time slot(s) for deferrable load appliances
according to an incentive price tariff from a utility company. The
HEMS 10 may communicate with the user via a wired interface or a
wireless interface. For example, the HEMS 10 may include a display
screen through which the user receives information from the HEMS 10
and an input device through which the user provides instructions to
the HEMS 10. In other implementations, the HEMS 10 may communicate
with the user via a smart phone application or other interface that
is wirelessly in communication with the HEMS 10.
Exemplary Learning-Based DR Strategy System
[0052] FIG. 4 illustrates a schematic diagram of a central server
500, or similar network entity, configured to implement a computer
system, according to one implementation. The server 500 executes
various functions of the HEMS 10 described above in relation to
FIGS. 2 and 3 and below in relation to FIGS. 5A and 5B. For
example, the server 500 may be the HAS 14 described above, or a
part thereof. As used herein, the designation "central" merely
serves to describe the common functionality the server provides for
multiple clients or other computing devices and does not require or
infer any centralized positioning of the server relative to other
computing devices. As may be understood from FIG. 4, the central
server 500 may include a processor 510 that communicates with other
elements within the central server 500 via a system interface or
bus 545. Also included in the central server 500 may be a display
device/input device 520 for receiving and displaying data. This
display device/input device 520 may be, for example, a keyboard or
pointing device that is used in combination with a monitor. The
central server 500 may further include memory 505, which may
include both read only memory (ROM) 535 and random access memory
(RAM) 530. The server's ROM 535 may be used to store a basic
input/output system 540 (BIOS), containing the basic routines that
help to transfer information across the one or more networks.
[0053] In addition, the central server 500 may include at least one
storage device 515, such as a hard disk drive, a floppy disk drive,
a CD Rom drive, or optical disk drive, for storing information on
various computer-readable media, such as a hard disk, a removable
magnetic disk, or a CD-ROM disk. As will be appreciated by one of
ordinary skill in the art, each of these storage devices 515 may be
connected to the system bus 545 by an appropriate interface. The
storage devices 515 and their associated computer-readable media
may provide nonvolatile storage for a central server. It is
important to note that the computer-readable media described above
could be replaced by any other type of computer-readable media
known in the art. Such media include, for example, magnetic
cassettes, flash memory cards and digital video disks.
[0054] A number of program modules may be stored by the various
storage devices and within RAM 530. Such program modules may
include an operating system 510 and a plurality of one or more (N)
modules 560. The modules 560 may control certain aspects of the
operation of the central server 500, with the assistance of the
processor 510 and the operating system 550. For example, the
modules may perform the functions described above in relation to
FIGS. 2 and 3 and below in relation to FIGS. 5A-6 and illustrated
by the figures and other materials disclosed herein, such as
executing various functions of the DR strategy system 10. According
to various implementations, one or more of the modules may be
executed by a digital computing system or portion thereof, such as
a micro computer, digital signal processor chip, field programmable
gate array (FPGA) system, PC, or other suitable computing
device.
[0055] In one exemplary implementation, the server 500 includes the
following modules: (1) a communication module for receiving data
relevant to modeling energy consumption and energy pricing and
communicating the DR strategy to the local receivers associated
with the energy consuming units; (2) a learning module for modeling
energy consumption based on relevant data using a computer
implemented learning-based approach; (3) an optimization module for
creating a next day DR strategy for regulatable loads based on DAP
data and the modeled energy consumption from the learning module
and for adjusting the DR strategy for the current day based on RTP
data for the current day; (4) an optimization module for creating a
next day DR strategy for fixed deferrable loads using the DAP or
RTP data and time windows; and (5) a data management module for
storing in queues in a memory actual energy consumption data,
temperature data (indoor and outdoor), and thermostat settings for
each time window for 1 through n days.
Learning Module
[0056] As noted above, the learning module models energy
consumption based on relevant data using a computer-implemented
learning-based approach. The learning-based approaches may include
a neural network approach or a regression-based approach, for
example. These exemplary approaches are described below beginning
with the description of FIG. 6. In addition, other suitable
learning mechanisms may be used. And, these learning mechanisms may
be used for modeling energy generation characteristics of a
renewable energy source, such as a PV system or a wind turbine
system, and/or energy storage characteristics of a battery for
storing energy generated by the renewable energy source.
[0057] Neural Network Approach
[0058] In a neural-network-learning-based approach, the model of
the energy consumed by the HVAC Q.sub.i is obtained through a
neural-network-based learning mechanism. Unlike the fixed model
shown by Equation 5, the HVAC energy consumption model learned by
the neural network is updated daily. Hence, it can more accurately
capture the thermal behavior of a house at different seasons,
users, weather conditions, etc. The neural network is trained by
using a backpropagation algorithm, which includes multiple
iterations until a stop criterion is reached. Thus, it is more
computational expensive.
[0059] One neural network learning approach is a multilayer
perceptron (MLP). A MLP is an artificial neural network structure
and may be useful in modeling HVAC energy consumption, for example.
As shown in FIG. 6, an MLP includes a set of source nodes that make
up the input layer, one or more layers of computation nodes, and an
output layer. The input signal propagates through the network in a
forward direction, on a layer-by-layer basis. The network exhibits
a high degree of connectivity, determined by the weights of the
network. Experiential knowledge for the network is acquired by the
network through a learning process and stored in the network
weights after it is trained.
[0060] For the learning purpose of HVAC energy consumption model,
the MLP 30 shown in FIG. 6 includes three input nodes 31a, 31b,
31c, one computation layer 33 that includes eight nodes, and one
output node 35. The three inputs nodes 31a, 31b, 31c of the network
30 receive the following data: 1) outside temperature T.sub.i.sup.O
in hour i (node 31a), 2) T.sub.i.sup.I room temperature in hour i
(node 31b), and 3) T.sub.i+1.sup.I room temperature in hour i+1
(also represents the thermostat setting temperature in hour i)
(node 31c). The network output is Q.sub.i, signifying the energy
consumed by the HVAC unit in hour i (FIG. 6). It is possible to add
other inputs to the network to enhance the learning.
[0061] As shown in FIG. 6, the network learning is based on the
data saved in four queues and is updated daily. The data saved in
the four queues are corresponding to outside temperature
(T.sub.i.sup.O), room temperature (T.sub.i.sup.I), thermostat
setting temperature (T.sub.i+1.sup.I), and HVAC energy consumption
(Q.sub.i).
[0062] A MLP network can be used for a function approximation
problem, in which the inputs to the network are equivalent to the
predictor variables as shown in the regression model of Equation 8
below and the output of the network is equivalent to the predicted
value. For a given problem, there is a cost function
.epsilon..sub.T, which is similar to the error sum of squares of
Equation 7 (shown below) for the regression model, as the measure
of training set learning performance. The objective of the learning
process is to adjust the weights of the network so as to minimize
.epsilon..sub.T. A highly popular training algorithm known as the
backpropagation algorithm is generally used to adjust the network
weights until a stop criterion is reached.
[0063] After the network is trained, the neural network would
provide a model that describes the relation of the HVAC energy
consumption (Q.sub.i) with the outside temperature (T.sub.i.sup.O),
room temperature (T.sub.i.sup.I), and thermostat setting
temperature (T.sub.i+1.sup.I) at a time frame i. This model is
updated daily and then used by the optimization module to determine
an optimal next-day DR strategy to operate the HVAC for the next 24
hours given the predicted outside temperature)(T.sub.i.sup.O and
electricity price.
[0064] Regression Based Approach
[0065] In a regression-learning-based approach, the model of the
energy consumed by the HVAC Q.sub.i is obtained through a
regression-based learning mechanism. Compared to the fixed model in
Equation 5, the regression model can also provide more accurate
estimation of HVAC energy consumption that is close to the actual
results, because the model is updated daily through the
regression-based learning mechanism. With regard to the
neural-network-based model, the regression-based learning is much
faster because the parameters of the regression model can be solved
directly.
[0066] Regression models quantitatively describe the variability
among the observations by partitioning an observation into two
parts. The first part of this decomposition is the predicted
portion having the characteristic that can be ascribed to all the
observations considered as a group. The remaining portion, called
the residual, is the difference between the observed value and the
predicted value and has to be ascribed to unknown sources. This can
be expressed as
y.sub.i=f(x.sub.i,.beta.)=.epsilon..sub.i i=1, . . . ,n (6)
where n is the number of the observations, y.sub.i is the ith
observation, x.sub.i=(x.sub.1, x.sub.2, . . . , x.sub.k) is the
predictor variable vector related to observation y.sub.i,
.beta.=(.beta..sub.0, .beta..sub.1, . . . , .beta..sub.p) is the
parameter vector, and .epsilon..sub.i is the error associated with
the ith observation.
[0067] The function f(.cndot.) is assumed to be smooth and
estimated by fitting a polynomial or other types of functions.
Fitting refers to calculating values of the parameters from a set
of data. Similar to the neural network learning, the estimate
{circumflex over (.beta.)}, a least squares estimate of .beta.,
tries to minimize the error sum of squares shown by.
S ( .beta. ^ ) = a n i = 1 ( y i - f ( x i , .beta. ^ ) ) 2 , ( 7 )
##EQU00002##
[0068] Regression approach is very effective if one knows the
general format of a function that the observations would follow.
Based on theoretical studies about building heat loads, a 3.sup.rd
order polynomial linear regression function is used to learn HVAC
energy consumption model as described mathematically by
Q.sub.j=q(T.sub.j+1.sup.I,T.sub.j.sup.I,T.sub.j.sup.O,.beta.) j=1,
. . . ,n (8)
where q(.cndot.) is a 3rd order polynomial function of
(T.sub.j+1.sup.I,T.sub.j.sup.I,T.sub.j.sup.O), n is the number of
the observations saved in the each of the queues as shown in FIG.
2, Q.sub.j is the jth HVAC energy consumption observation,
(T.sub.j+1.sup.I,T.sub.j.sup.I,T.sub.j.sup.O) is the predictor
variable vector consisting of thermostat settings and indoor and
outdoor temperatures related to observation Q.sub.j, and
.beta.=(.beta..sub.1, .beta..sub.1, . . . , .beta..sub.p) is the
parameter vector, wherein p represents the total number of
coefficients of the polynomial function. It is possible to add
other inputs to the predictor variable vector to enhance the
learning. Similar to the neural network approach, the regression
model is updated daily when new data is saved in the queues each
day. Compared to the simplified model Equation 5, the regression
model can also provide more accurate estimation of HVAC energy
consumption that is close to the actual results.
[0069] Similar to the neural network, the regression approach would
provide a model that describes the relation of the HVAC energy
consumption (Q.sub.i) with the outside temperature (T.sub.i.sup.O),
room temperature (T.sub.i.sup.I), and thermostat setting
temperature (T.sub.i+1.sup.I) at a time frame i. This model is
updated daily and is then used by the optimization module to
determine an optimal next-day DR strategy to operate the HVAC for
the next 24 hours given the predicted outside
temperature)(T.sub.i.sup.O) and electricity price.
[0070] Model Based Approach
[0071] As discussed above, the energy consumption model of the HVAC
can be obtained by using a learning based approach, such as the
neural network approach or regression based approach described
above. In addition, the learning module may also model energy
consumption by using a model-based approach. In the model based
approach, the energy consumed by the HVAC Q.sub.i is modeled as a
function of room temperature T.sub.i.sup.I, room temperature
T.sub.i+1.sup.I in hour i+1 (also represents the thermostat setting
temperature in hour i), and outdoor temperature T.sub.i.sup.O.
Equation 5 above shows an example of a simplified thermal model of
a residential house. However, in reality, the relationship between
indoor and outdoor temperatures and energy consumed by a HVAC unit
is much more complicated than Equation 5 so that actual energy
consumption of a HVAC unit could deviate greatly from results
generated by using Equation 5, which would affect the DR
efficiency.
Optimization Modules
[0072] The optimization modules generate DR strategies for one or
more energy units having regulatable loads and/or fixed deferrable
loads. In particular, the optimization module for creating DR
strategies for regulatable loads may solve the optimization problem
of Equation 4 to create a DR strategy for each regulatable energy
unit to minimize the costs of energy consumption for a particular
day. For example, the optimization module may use one of the
following optimization approaches to solve the optimization
problem: (1) a nonlinear or linear programming approach and 2)
particle swarm technique. These are discussed in more detail below
in relation to FIGS. 5A and 5B and Algorithms 1 and 2.
[0073] The optimization module for creating DR strategies for fixed
deferrable loads may generate a DR strategy for each fixed
deferrable load that may be based on real time energy pricing,
acceptable price ranges, and/or time frames during which these
energy consuming units may be operated. The energy usage of fixed
deferrable loads is considered to have a fixed energy consumption
pattern. The optimization module may use a binary integer
programming technique to solve the optimization problem for fixed
deferrable loads.
[0074] If assuming there are one HVAC load, one deferrable load,
and one future fixed load, the robust integrative optimization can
be described mathematically by
Minimize : C = i ( p i - p i PV - p i batt ) ( Q i HVAC + Q i
deferrable + Q i fixed ) i = 1 , , 24 Subject to : 0 .ltoreq. Q i
HVAC .ltoreq. Q max , T min .ltoreq. T i + 1 I .ltoreq. T max , Q i
HVAC = q ( T i + 1 I , T i I , T i O , w .fwdarw. ) i = 1 , , 24 (
9 ) ##EQU00003##
in which p.sub.i.sup.PV and p.sub.i.sup.batt are day-ahead dynamic
discount price tariffs associated with PV and battery. The purpose
of the discount price tariffs is to generate a virtual low price
signal to home appliances so as to shift DR appliances to those
"low price" time frames, such as during a high PV generation time
period during day time or during a battery discharging time period
after sunset. Similarly, computer-based learning systems may be
configured for formulating dynamic discount price tariffs
associated with PV power production and battery storage capability
at different seasons, weather, and house conditions. Then, the next
step is to solve the overall optimization problem to determine an
optimal DR policy to operate the home appliances. Furthermore, the
system may also generate a control and management strategy for the
battery for the next day based at least on a charge of the battery
for the current day and an expected, or modeled, difference between
next-day energy generation from at least one renewable energy
generation source and energy consumption from at least one energy
consuming unit. The control and management strategy may include one
or more time windows during which the battery is charged when
energy generation is higher than energy demand or when energy from
the battery is discharged for consumption by one or more energy
consuming units when energy generation is not occurring or is
occurring at a level below a predetermined acceptable
threshold.
[0075] To solve the integrative optimization problem, many previous
studies focused on developing methods to predict energy consumption
patterns of all home appliances, including fixed deferrable and
fixed non-deferrable loads, to manage the loads in the DR
framework. However, due to various factors, the forecast data are
far from accurate. Thus, most of these solutions used stochastic
optimization techniques to solve the optimization problem. However,
inventors discovered that knowing energy consumption patterns for
fixed deferrable and fixed non-deferrable is not needed in
optimization formulation. Instead, inventors have decoupled the
complicated and multi-objective optimization problem into small,
independent optimization sub-problems, which allows the
optimization modules to generate DR policies for regulatable and
fixed deferrable loads more effectively and efficiently.
[0076] In particular, in Equation 10 below, fixed loads are assumed
to be future constant loads although it may not be known exactly
how much electric energy a fixed load will use for the next day.
This future fixed load concept may be applied to energy generated
from renewable sources as well (such as solar PVs or wind turbines)
except that the energy consumption from a renewable source is
negative instead of positive. Hence, removing the fixed loads and
fixed renewable sources from the objective function does not affect
the optimal solution so that an equivalent and simpler optimization
problem may be obtained, which is described by
Minimize : C = i ( p i - p i PV - p i batt ) ( Q i HVAC + Q i
deferrable ) Subject to : 0 .ltoreq. Q i HVAC .ltoreq. Q max , T
min .ltoreq. T i I .ltoreq. T max ( 10 ) ##EQU00004##
Also, unlike the HVAC load, a fixed deferrable load has a fixed
energy consumption pattern. This fixed load pattern can be applied
to different time frames. Assuming that the fixed load pattern is
Q.sup.deferrable within one-hour time frame, then Equation 10 can
be simplified as
{ Minimize : C HVAC = i ( p i - p i PV - p i batt ) Q i HVAC
Subject to : 0 .ltoreq. Q i HVAC .ltoreq. Q max , T min .ltoreq. T
i .ltoreq. T max + { Minimize : C deferrable = i ( p i - p i PV - p
i batt ) .theta. i Q deferrable ( 11 ) ##EQU00005##
where .theta..sub.i is a binary integer array to indicate the
status of the deferrable load (e.g., on or off). For example, if
the deferrable load is operated at ith hour of the day for one
hour, then,
.theta. i = ( 0 0 1 ith hour 0 0 ) T . ##EQU00006##
Equation 11 indicates that the integrative optimization problem of
Equation 10 is actually equivalent to two independent optimization
problems. In summary, the optimization and energy management
problem for fixed non-deferrable loads, regulatable loads, and
fixed deferrable loads can be solved separately and then combined
together to achieve the final optimal solution. Note that loads in
each category are also independent. Therefore, a complicated and
multi-objective optimization problem is decoupled into several
small independent optimization problems that are used to generate
robust optimal energy management policy and decisions effectively
and efficiently. Energy consuming units having future fixed
non-deferrable loads may be turned off via one or more sensors when
not needed. For example, as shown in FIG. 20, lights in a room may
be turned on in response to one or more sensors detecting the
presence of a person in the room and turned off in response to one
or more sensors detecting that the person has left the room (either
via no input to that sensor or via input received by a sensor in
another room).
[0077] The DR strategy for regulatable loads may be based on the
modeled next-day energy consumption and day-ahead energy pricing.
For example, for the HVAC unit, the optimization module uses the
HVAC energy consumption model developed by the learning module to
estimate HVAC energy consumption in each time window for the next
day (e.g., 24 hours) and creates a DR strategy for the HVAC unit
based on the modeled consumption, the weather forecast, and the
next day energy prices. The DR strategy for the HVAC unit includes
a plurality of thermostat settings corresponding to each time
window for the next day that are the most cost-effective settings
for the next day. Because the model is updated daily, an optimal
and substantially real-time DR strategy that is solved according to
the learning-based adaptive model results in more efficient
operation of a HVAC unit for different seasons, users, and weather
conditions. The optimization module may also be configured for
adjusting the DR strategy for the current day based on RTP data for
the current day. Initially, the queues used by the learning module
do not contain sufficient amounts of data for the learning module
to use a learning based approach to model energy consumption. Until
the queues are sufficiently full, the optimization module may use
an optimization approach based on a simplified energy consumption
model, such as shown in Equation 5, and DAP data or a heuristic
approach (see FIG. 1) and RTP data to generate a DR strategy for
one or more energy consuming units.
[0078] For fixed deferrable loads, such as the dishwasher or dryer,
the DR strategy may be based on real time energy pricing and
acceptable price ranges and/or time frames during which the energy
consuming unit may be operated. For these energy consuming units,
the optimization module may receive from the customer an indication
that the unit is operable when the RTP is lower than a predefined
price set by the customer. This indication may be subject to
operation during an acceptable time frame. Otherwise, those energy
consuming units operate at the time frame that has the lowest price
within a customer preferred time window. By delaying the operation
of these energy consuming units, the total energy consumption of
the house does not change. However, delaying operation to a
non-peak or lower price time period may reduce the peak load and
decrease the overall energy cost for the day, depending on the
prices for the day.
[0079] For example, when DR strategies are not used, it may be
assumed that the dishwasher operates between 7 pm to 8 pm and the
dryer operates between 6 pm to 9 pm on Monday, Wednesday, and
Friday. On Sunday, 50% usage of dishwasher and dryer is defined
between 7 pm to 8 pm and 6 pm to 9 pm, respectively, for
consideration of general light load usage of those appliances
during the weekend.
[0080] Using a DR strategy, the customer may specify instead that
the dishwasher may operate for an hour or less one time every two
days, for example, but the customer may not have a preference as to
when it is operated so long as the price of electricity during
operation is within a particular range set by the customer. The
operation schedule and/or preferred pricing ranges may be input by
or at the direction of the customer, for example. In other
implementations, the usage schedule may be set by another entity,
such as the utility company, or set by the HEMS 10 based on DTP,
RTP, energy generated by renewable sources associated with the
house, and/or expected usage of other energy consuming units in the
house.
Exemplary Flow of the HEMS 10
[0081] FIGS. 5A and 5B illustrate an exemplary flow 100 of the HEMS
10 for generating a DR strategy for an HVAC unit according to one
implementation. If actual data is not available upon start up of
the system 10, the system 10 establishes an initial DR strategy to
use until sufficient initial data is collected and stored in the
queues 26. Steps 101 through 113 below describe exemplary steps for
establishing an initial DR strategy until sufficient amounts of
actual measured data are available. The system 10 may use a
simplified DR strategy, such as a heuristic DR method (FIG. 1), to
run the HVAC unit until sufficient measured energy consumption data
of the HVAC unit is stored in the queues 26. For a new home or HVAC
unit, the data queues are empty at the beginning, so an
optimization routine cannot be developed and applied. Thus, initial
DR policies may be generated by using a simplified DR
mechanism.
[0082] Steps 115 through 127 describe exemplary steps for
establishing a more accurate DR strategy once sufficient data is
stored. After the data queues 26 are full, the learning module
learns the HVAC energy consumption model based on the measured
energy consumption data stored in the queues 26, and the
optimization module generates optimal DR policy everyday based on
the learned model, weather forecast, and forecasted electricity
pricing (DAP). The learning and optimization modules may be
implemented via a substantially real-time computing system, such as
digital signal processing (DSP) chips, by using
assembly-language-based software.
[0083] Referring back to FIG. 5A, beginning at Step 101, DAP data
and weather forecast for a location of the building are received by
the system 50, such as by the communication module. For example,
24-hour electricity DAP data can be provided by an electric utility
one day ahead under a dynamic pricing framework. The energy pricing
data for electricity is typically provided by the electric utility
company. Energy pricing data for natural gas is typically provided
by the natural gas company. In addition, 24-hour weather forecast
data may be available from National Weather Service, for example.
The weather forecast data for the location may include 24-hour
next-day weather forecast data and current-day weather data, such
as predicted and current hourly temperatures, humidity, and wind
speed and direction.
[0084] In Step 103, an estimated usage schedule for the HVAC unit
is received. The estimated usage schedule includes, for example,
when and for how long the HVAC is operated and how much energy is
needed to operate it. For example, the HVAC unit may be operated
throughout the day depending on the indoor temperature, and the
length of time for operation and the amount of energy needed for
operation depends on the outdoor temperature, the thermal
characteristics of the house, and the efficiency of the HVAC
unit.
[0085] A next-day DR strategy is created in Step 107 by using a
simplified DR technique, such as 1) an optimal HVAC DR policy
generated based on the simplified energy consumption model
(Equation 5) and DAP data or 2) an HVAC DR policy generated by
using a heuristic approach based on DAP data (FIG. 1). The next-day
DR strategy sets forth the operating parameters of one or more
energy consuming units at different times of the day. For example,
the DR strategy may set forth the hourly thermostat settings for
the HVAC system and time frames for operation of the dishwasher,
washer, and/or dryer.
[0086] Referring back to the implementation shown in FIG. 5A, if
RTP data is available for the location, the RTP data may be
received in Step 108, and the heuristic approach or the simplified
optimization approach may be used with the next-day DR strategy and
RTP data to adjust the DR strategy during the current day in which
the DR strategy is applied to optimize the energy costs for the
customer, as shown in Step 109. This step may be omitted, however.
For example, Steps 108 and 109 may be omitted if RTP data is not
available for the location or if there are other incentives given
to customers by the utility company. In an alternative
implementation (not shown), the RTP data may be used to evaluate
the effectiveness of the DR strategy for the current day in
addition to or in lieu of Step 109. Actual energy consumption data,
indoor temperatures, outdoor temperatures, and thermostat settings
for each time window (e.g., each hour or preset time frame) during
a day are stored in the memory in respective queues 26, as shown in
Step 111, by the data management module. The data is stored in each
queue on a FIFO basis for a certain number of days n. For example,
in one implementation, the data may be stored for 8 days. Thus,
when new data is received each day, the oldest data saved in the
front of the queues is removed and the new data is saved.
Maintaining the queue on a FIFO basis provides up-to-date household
energy consumption data for the learning module, which allows the
learning module to more accurately capture the thermal and
appliance energy usage behavior of a house for different seasons,
users, and weather conditions, for example. The number of days for
which the queues are configured for storing data may be selected
such that any update of the HVAC energy consumption based on the
data saved in the queues should reflect the impact of the seasons,
weather, users, and house conditions, for example, within one or
two weeks.
[0087] The system repeats Steps 101 through 111 until the memory
queues 26 are full, as shown in Step 113. Alternatively, if a data
acquisition or storage component of the system 10 has a problem
that prevents storage of data for one or more days into the queues
26, Steps 101 through 113 may be executed until the model is
stabilized (e.g., the queues 26 are fully populated by data that
can be used by the learning module).
[0088] Once the memory queues 26 are full, DAP data and weather
forecast data for the location are received in Step 115, such as by
the communication module. Then, in Step 117, the learning module
models the energy consumption for the next day using a
learning-based approach, such as neural network learning or
regression-based learning, which are described above in the section
entitled "Learning Module." The modeled energy consumption is based
on the weather forecast data and actual usage data stored in the
memory queues 26, for example. In one implementation (not shown),
the learning module may be configured for receiving a selection
from the user regarding which learning based approach the user
prefers for the system 10 to use. In another implementation (not
shown), the user may be able to select a particular learning based
approach for the system 10 prior to or during installation of the
system 10.
[0089] In Step 119, a DR strategy is created for the next day by
the optimization module based on the energy consumption modeled in
Step 117 and the DAP data received in Step 115. This DR strategy is
communicated to the HVAC unit in Step 126 by the communication
module.
[0090] In Step 127, RTP data and energy usage data for the current
day may be received by the communication module, and in Step 121,
the DR strategy that was created the day prior for the current day
may be adjusted during the current day based on the usage data
and/or RTP data received for the current day. However, if RTP data
is not available for the location or if there are other incentives
given to customers by the utility company, these steps may be
omitted.
[0091] In Step 123, the data management module removes usage data
from the queues for the oldest day and stores usage data for the
most recent day (e.g., the current day or one day prior) in the
queues. Then, Steps 115 through 127 are repeated for the next
day.
[0092] The various modules described above in FIGS. 4 and 5A-5B are
exemplary and the functions described as being performed by each
module may be performed by other modules executed on the same or
another computing device.
[0093] Except for the queues, it is possible to use other data
structures, such as trees, tables, or database management systems,
to store the weather and usage data.
[0094] FIGS. 5A and 5B illustrate an exemplary process for
generating a DR strategy for the HVAC unit, but the process and
data described above may be adapted for other types of regulatable
energy units.
[0095] Algorithm 1 below presents an exemplary pseudo-code that
illustrates how an HVAC energy consumption model is learned and how
an optimal DR policy for operating the HVAC unit is generated and
updated. Lines 1 to 8 correspond to Steps 101 through 113 of FIG.
5A for establishing an initial DR strategy until sufficient amounts
of actual data are available.
[0096] Lines 9 and 10 correspond to Steps 115 through 127 of FIG.
5B for establishing a more accurate DR strategy once sufficient
data is stored. A HVAC energy consumption model for the time window
represented by the length of the queues is obtained through the
regression or neural network approach, based on which an optimal DR
policy is generated (line 10). The DR policy is loaded into a
programmable thermostat to control HVAC energy consumption for the
next day. During the operation of the HVAC unit the next day, the
actual measured energy consumption results, together with the
actual measured weather information, 24-hour thermostat settings
and actual measured 24-hour indoor room temperature, are saved in
the queues. At the same time, old 24-hour data in the front of the
queues are removed. The process continues to update HVAC energy
consumption model and generate new optimal DR policy day after
day.
TABLE-US-00001 Algorithm 1: Learning-based demand response 1:
{Initial demand response Steps 101 through 113as shown in FIG. 5A}
2: for d=1 to numDay 3: T.sub.queue.sup.O .rarw. T.sub.i.sup.O ,
P.sub.queue .rarw. P.sub.i (i=1,......, 24) 4: Obtain 24-hour
thermostat settingT.sub.i(i=1, ..., 24) (Step 107 as shown by Fig.
5A). 5: Programmable thermostat T.sub.i(i=1, ..., 24). 6: Obtain
next-day HVAC energy consumption Q.sub.i(i=1, ..., 24) through
real-time measurement 7: Q.sub.queue .rarw. Q.sub.i, T.sub.queue
.rarw. T.sub.i(i=1, ..., 24) 8: end for 9: {Updating HVAC energy
consumption model and generating optimal DR policy (Steps 115
through 127 as shown by FIG. 5B) using nonlinear programming or PSO
method (see discussion of optimization module below) } 10: Running
algorithm day after day (go back to line 9).
[0097] Algorithm 2 shown below provides the pseudo-code of the
particle swarm optimization (PSO) method according to one
implementation. In PSO, each single solution is called a particle
in the search space. All of particles have fitness values which are
evaluated by the fitness function to be optimized, and have
velocities which direct the flying of the particles. Each candidate
solution can be thought of as a particle "flying" through the
fitness space by updating the candidate velocity and position based
on both global best fitness position and local best fitness
position.
TABLE-US-00002 Algorithm 2: PSO during the iteration of Alg. 1 in
line 9 1: Initialize particle position: T.sub.buf (m) .di-elect
cons. [T.sup.min,T.sup.max ] 2: Initialize particle speed: {tilde
over (T)}.sub.buf(m) .di-elect cons.[-.DELTA.T,.DELTA.T], m=1,...,M
3: {Calculate initial fitness values for all particles}
fitness(m)=f(T.sub.buf(m)), m = 1,...,M 4: {circumflex over
(T)}.sub.G .rarw. T(k); if fitness(k)=max{fitness(m),m.di-elect
cons. [1, M]} {circumflex over (T)}(m) .rarw. T(m); 5: do 6:
{Update velocity for all particles} {tilde over (T)}(m) = w{tilde
over (T)}.sub.buf(m)+c.sub.1 rand(0,1)[{circumflex over
(T)}(m)-T.sub.buf(m)] +c2 rand(0,1)[{circumflex over
(T)}.sub.G-T.sub.buf (m)] 7: {Update position for all particles}
T(m) = T.sub.buf (m) + {circumflex over (T)}(m) 8: if T(m) out of
boundary .fwdarw. boundary handling T(m) 9: {Calculate fitness
values for all particles} fitness(m)=f (T(m)), m = 1,... ,M 10:
{circumflex over (T)}.sub.G .rarw. T(k); if
fitness(k)=max{fitness(m), m .di-elect cons. [1,M]} {circumflex
over (T)}(m) .rarw. T(m); if T(m) > T.sub.buf(m) 11:
T.sub.buf(m).rarw.T(m); {tilde over (T)}.sub.buf(m).rarw.{tilde
over (T)}(m) 12: while maximum iterations or a stop criteria is not
reached 13: Output global optimal solution {circumflex over
(T)}.sub.G to Alg. 1.
[0098] In Alg. 2, T(m), the position of a particle, represents a
24-hour thermostat setting during a day. {tilde over (T)}(m), the
speed of a particle, represents the thermostat adjustment during
one iteration of the PSO algorithm. {circumflex over (T)}(m) and
{circumflex over (T)}.sub.G are the individual best thermostat
setting associated with particle m and the global best thermostat
setting of all the particles, respectively. In line 6, c.sub.1 and
c.sub.2 are the user defined coefficients, and w is the inertia
weight used to balance global and local search. These parameters
are determined through trial and error until a best possible result
is obtained. For each updated particle position calculated in line
7, it is checked in line 8 whether the updated position is out of
the boundary. If so, for any temperature setting beyond [T.sup.min,
T.sup.max] it is reset to T.sup.min or T.sup.max depending on
whether the temperature setting is smaller than T.sup.min or larger
than T.sup.max [40]. The fitness function f(.cndot.) is defined
by
f ( T m ) = { 1 / [ i = 1 24 p i Q ( m ) i ] , Q min .ltoreq. Q ( m
) i .ltoreq. Q max 0 else ( 12 ) ##EQU00007##
where Q(m).sub.i=q(T.sub.i+1,T.sub.i,T.sub.i.sup.O,.beta.) is the
energy consumption of the HVAC for particle m at ith hour (i=1, . .
. , 24), and Q.sup.min and Q.sup.max stand for the minimum and
maximum HVAC energy consumption corresponding to a practical HVAC
unit.
[0099] The global market for grid-connected residential solar PV
installations coupled with energy storage is predicted to grow
tenfold to reach more than 900 megawatts in 2018, making smart
management and control of residential PV systems and energy storage
increasingly important. To electric utilities, proper management of
PV and energy storage will help to maintain grid reliability,
resiliency, and power quality while minimizing curtailment of
available solar power. To home owners, effective use of PV and
energy storage will help consumers to save money off the
electricity bill. Residential PV and energy storage that is managed
and controlled holistically according to dynamic electricity prices
and tariffs in a demand response (DR) framework may benefit the
utility company and residents. Based on this framework, the HEMS
provides the most efficient and cost effective grid integration of
residential energy storage, solar PVs, and home appliances,
according to various implementations.
[0100] For example, FIGS. 16 and 17 illustrate diagrams of a
residence in which solar PVs, home appliances, and residential
energy storage are integrated into the HEMS 10. In particular, the
HEMS 10 according to the implementation shown in FIG. 17 further
includes an inverter 4, a battery 7 for storing energy generated by
the PV system 1 and discharging energy to energy consuming units
within the building, a supercapacitor in electrical communication
with the battery 7 that absorbs most of the large and fast ramps in
PV power output (e.g., caused by cloud movement). The PV system 1
may be configured for operating in maximum power point tracking
(MPPT) mode, according to one implementation. The HAS 14, or
another controller in electrical communication with the HEMS 10,
may be further configured for: (1) using artificial neural networks
(ANNs) and adaptive dynamic programming (ADP) to control the
inverter 4, (2) identifying when power should be released or stored
by the battery by integrating real-time dynamic impedance
measurements for battery charging/discharging algorithms, and (4)
coordinating usage of regulatable and deferrable energy consuming
units with PV and energy storage.
Security System for HEMS
[0101] Various implementations of the HEMS 10 may also include a
security system that assures that DR policies generated one day or
hours ahead are executed next day or later without modification.
The security system may communicate with a smart meter that is
configured for proving the correctness of any smart appliance, a
group of smart appliances that are configured for proving the
correctness of the smart meter, and a service provider configured
for proving the correctness of the smart meter. In one exemplary
security system, each meter has a copy of its own log, and each
household meter is mapped to several other witnesses. The system
also includes a commitment protocol that is used to ensure that
witnesses will retrieve exactly the same log as the observation
object owns. In addition, the system includes a challenge/response
protocol to address the problem that some household meters do not
response or fail to acknowledge whether messages were sent
successfully or not.
[0102] The security system maintains the confidentiality,
availability, and integrity of information and systems and methods
of implementing the above-described functionality and other smart
home functions. In addition, accountability mechanisms may be also
be included that provide trustworthy smart appliances in homes and
overcome the challenges for integration of physical systems and
human factors. Thus, the security system may be configured for (1)
accountable, non-repudiable, malicious appliance inspection, (2)
denial of service (DoS) attack mitigation for operation of diverse
DR appliances and renewable sources in the smart home area, and (3)
interoperability of energy system, wireless communication, and
security methods.
[0103] The accountability mechanisms may be configured for
verifying the customer responses and the smart appliance group and
for adopting logic to provide theoretical proofs of accountability
for the designed protocols. The security mechanisms and protocols
may also allow a utility company to prove the correctness of DR
policies actually used to operate home appliances.
[0104] The designed protocols may achieve the following exemplary
goals: (a) a smart meter that can prove the correctness of smart
appliances in a home; (b) a group of smart appliances that can
prove the correctness of the smart meter; and (c) a service
provider that can prove the correctness of the smart meter. The
witness mechanisms may include several protocols. For example, one
protocol may be that every meter has one copy of its own log, which
is ensured by a tamper-evident, log mechanism. Other logs are
retrieved when required. Meters exchange just enough messages to
prove themselves. As another example, each household meter is
mapped to several other witnesses. Each witness collects its log,
check its correctness by comparing the readings, and reports the
results to the rest of the system. The witnesses are reassigned
observation objects according to function w at set intervals.
Another example includes a commitment protocol that is used to
ensure that witnesses will retrieve exactly the same log, as the
observation object owns. It also guarantees that no one can deny a
received message. Furthermore, another example includes a
challenge/response protocol to address the problem that some
household meters do not respond or fail to acknowledge whether
messages were successfully sent or not.
[0105] Mutual-inspection and malicious appliance inspection
mechanisms may also be included. Mutual-inspection mechanisms may
realize non-repudiation in smart homes so that any misbehavior and
malicious operation compromising power reading is eventually
detected either by the utility company or by independent users. The
system may also jamming mitigation techniques, sampling
synchronization, and how to prevent wireless network delay and
wireless jamming. The malicious appliance inspection problem may be
addressed via a suite of tree-based inspection algorithms
(including an adaptive tree approach) for both static inspection
and dynamic inspection, which are investigated through theoretical
analysis, simulations, and experiments.
[0106] Second, to prevent the HEMS wireless networks from being
jammed maliciously to cause DoS attacks, the security system may
include DoS attack mitigation methods for diverse DR appliances and
renewable sources in the smart home area. For example, these
methods may include channel hopping when a DoS attack happens, such
as rendezvous channel hopping algorithms via catalan number series,
as channel hopping patterns with adjustments.
[0107] Third, interoperability and security are tightly coupled
activities, and these joint activities satisfy multiple
requirements in both domains, including interoperability
requirements for transparency and clarity in system interfaces and
relationships, and security requirements for resistance to attack,
self-checking, system access control, etc.
Computational Experiment System for Comparing DR Strategies Based
on Various Energy Consumption Models
[0108] A great difficulty for DR study is how to investigate and
evaluate the performance of different DR strategies because such
study requires the DR strategies to be applied to the same house
condition. For a practical residential house, it is impractical to
meet this requirement.
[0109] DR strategies developed using various learning-based
modeling approaches may be compared, researched, and evaluated by
using a computational experiment system that simulates energy
consumption of a real-life house. This section describes an
exemplary computational experiment system for comparing DR
strategies based on various energy consumption models. The
computational experiment system 60 shown in FIGS. 7 and 9 is
similar to the real-life DR strategy system 10 described above in
relation to FIGS. 2-5B except that the system 60 uses simulated
energy consumption data generated by building simulation software
instead of actual energy consumption data, according to various
implementations. In addition, the system 60 may generate multiple
DR strategies based on various energy consumption modeling
approaches so that the cost and energy effectiveness of the
multiple DR strategies may be compared. In one implementation, this
system 60 may provide information to the end user comparing the
energy costs and/or consumption resulting from use of the DR
strategy system 10 and the costs and consumption resulting from
another type of DR system or from not using a DR strategy.
[0110] To accomplish the computational experiment system, MATLAB, a
powerful technical software system, is integrated with building
simulation software for the DR evaluation, according to one
implementation. The integrated simulation strategy may greatly
accelerate the offline design and assessment of a DR strategy.
Other suitable software can also be used for such offline
simulation evaluation.
[0111] The building simulation software, such as eQUEST, Energy
Plus and others, is used as a virtual test bed to simulate home
energy consumption. The building simulation software allows one to
"build" a simulated house that is similar to a practical one. It
uses standard commercial building materials defined in the software
library and real-life weather and solar data, which may be obtained
from the US Department of Energy, for example. Thus, the building
simulation software is capable of estimating changes in the
electrical load of a "practical" house throughout a year, certain
days within the year, or certain time period of a day.
[0112] To simulate energy consumption of a residential house, an
architectural model of the house is created based on the blueprint
and construction materials used to build the house. For example, a
generic floor plan for a two story, 2,500 square foot house may be
used with a location selection of Springfield, Ill. Next, internal
loads are defined. The internal loads may include, for example,
washing machines, dryer, water heater, dishwasher, electric stove,
refrigerator, microwave oven, lights for each room, occupants, etc.
Then, appropriate schedules are input to model these loads, such as
when, how long, and how much of these loads are used each day. The
final step is defining the HVAC system. In this example, a
three-ton unit is selected to condition the first floor and a
two-ton unit is selected to condition the second floor. Additional
details about how to build a simulated house can be found in the
eQuest Introductory Tutorial, version 3.64 by James J. Hirsch &
Associates, December 2010, for example. The simulation software
uses the above information to simulate the amount of energy
consumed during various time windows of a day (e.g., hourly) based
on the actual weather data for that day for the specified
location.
[0113] FIG. 7 illustrates a schematic diagram of a central server
600 for implementing a computational experiment system 60 according
to one implementation, and FIGS. 8A-9 illustrate an exemplary flow
for the system 60 according to one implementation.
[0114] As shown in FIG. 7, the central server 600 may include a
processor 610 that communicates with other elements within the
central server 600 via a system interface or bus 645. Also included
in the central server 600 may be a display device/input device 620
for receiving and displaying data. This display device/input device
620 may be, for example, a keyboard or pointing device that is used
in combination with a monitor. The central server 600 may further
include memory 605, which may include both read only memory (ROM)
635 and random access memory (RAM) 630. The server's ROM 635 may be
used to store a basic input/output system 640 (BIOS), containing
the basic routines that help to transfer information across the one
or more networks.
[0115] In addition, the central server 600 may include at least one
storage device 615, such as a hard disk drive, a floppy disk drive,
a CD Rom drive, or optical disk drive, for storing information on
various computer-readable media, such as a hard disk, a removable
magnetic disk, or a CD-ROM disk. As will be appreciated by one of
ordinary skill in the art, each of these storage devices 615 may be
connected to the system bus 645 by an appropriate interface. The
storage devices 615 and their associated computer-readable media
may provide nonvolatile storage for a central server. It is
important to note that the computer-readable media described above
could be replaced by any other type of computer-readable media
known in the art. Such media include, for example, magnetic
cassettes, flash memory cards and digital video disks.
[0116] A number of program modules may be stored by the various
storage devices and within RAM 630. Such program modules may
include an operating system 650 and a plurality of one or more (N)
modules 660. The modules 660 may control certain aspects of the
operation of the central server 600, with the assistance of the
processor 610 and the operating system 650. For example, the
modules may perform the same or similar functions described above
in relation to FIGS. 4 and 5A-5B and below in relation to FIGS.
8A-8B and illustrated by the figures and other materials disclosed
herein, such as executing various functions of the computational
experiment system 60.
[0117] In one exemplary implementation, the server 600 includes the
following modules: a simulation module, a learning module, an
optimization module, a data management module, and a pricing
module.
[0118] The simulation module uses the next-day DR strategy
determined the day before by the optimization module for the system
60 and current day weather data to simulate the energy consumption
of the house for the current day. In particular, energy consumption
of a residential house is simulated by using the simulation module
of the computational experiment system 60 for a practical weather
pattern of the day, including temperature, humidity, solar
radiation, etc., at a location. The results generated by the
simulation module of system 60 are stored in the queues and are
received by the learning module of system 60 to update the modeled
energy consumption each day.
[0119] The learning, optimization, and data management modules of
the computational experiment system 60 are similar to the
corresponding modules described above for the DR strategy system
10, except that the queues include simulated energy consumption
data generated by the simulation module instead of actual energy
consumption data. FIG. 9 illustrates the schematic flow of data
through the queues and the learning, optimization, and simulation
modules of the system 60.
[0120] The pricing module calculates electricity pricing for the
current day based on RTP data and the simulated energy consumption
for the house. The pricing module may also calculate energy pricing
for the next day based on DAP data and the simulated energy
consumption for the next day. For an implementation in which two or
more learning approaches are used to model the energy consumption
of the house and a DR strategy is determined based on each learning
approach, the pricing module may generate one or more displays
indicating the calculated prices for energy usage over the course
of a day using each DR strategy, such as the graphs shown in FIGS.
11-15.
[0121] FIGS. 8A-8B illustrate an exemplary flow 200 of the
computational experiment system 60 according to one implementation.
Beginning at Step 201 in FIG. 8A, the DAP pricing data and weather
forecast data are received for a particular location. In Step 203,
the estimated usage schedule for an HVAC unit for the house for the
next day is received. In Step 205, a next-day DR strategy is
created by the optimization module by using a simplified DR
technique, such as (1) the optimal HVAC DR approach based on the
simplified energy consumption model shown in Equation 5 and DAP
data or (2) a heuristic approach based on DAP data (FIG. 1). Then,
in Step 207, the DR strategy is communicated to the simulation
module to simulate energy consumption for the next day based on the
DR strategy and actual weather for that day. If RTP data is
available for the location, the RTP data and the simulated energy
usage for the HVAC unit may be received for the current day, as
shown in Step 208, and the RTP is used to adjust the DR strategy
during the day it is being applied, as shown in Step 211.
[0122] In Step 209, the simulated energy consumption data, the
thermostat settings, and the temperature data (indoor and outdoor)
for each time window for the current day are stored in the memory
queues.
[0123] In Step 213, the system 60 determines whether the queues are
full. If the queues are not full, then Steps 201 through 211 are
repeated for the next day. If the queues are full, then the system
60 proceeds to Step 215 in FIG. 8B in which DAP and weather
forecast data are received for the next day. In Step 217, the
learning module generates at least one energy consumption model for
the HVAC unit using one or more learning-based approaches based on
the weather forecast, thermostat settings, and the simulated energy
consumption data stored in the queues. In Step 219, the
optimization module generates one or more DR strategies based on
the DAP, the weather forecast, and the one or more energy
consumption models from Step 217. The DR strategies may be based on
the same modeling approach or different modeling approaches.
Alternatively, or in addition thereto, the DR strategies may be
solved using the heuristic approach, the optimal approach from
Equation 5 or Equation 8, or the PSO approach using Equation 4 or
Equation 8. In Step 221, the DR strategies are communicated to the
simulation module to simulate energy consumption for the next day
based on the actual weather. In Step 223, the simulated energy data
from the simulation module is communicated to the data management
module for storing the data in the queue, and old data is removed.
In addition, in Step 225, RTP data, if available, is received along
with simulated energy usage data for the HVAC unit for the current
day. Then, in Step 227, the RTP data is used to adjust the DR
strategy during the day it is applied. Steps 215 through 227 are
then repeated each day. The energy costs and/or usage associated
with each DR strategy are displayed by the pricing module (not
shown), which allows a user to compare DR strategies and which
yield lower energy costs and usage.
[0124] FIGS. 8A and 8B illustrate an exemplary process for
generating one or more DR strategies for the HVAC unit, but the
process and data described above may be adapted for other types of
energy consuming units.
[0125] The exemplary pseudo code in Alg. 1 above may be adapted for
use with the simulated energy consumption data from the simulation
module to model energy consumption and generate a DR policy for the
next day. Lines 1 to 8 correspond to Steps 201 through 213 of FIG.
8A for establishing an initial DR strategy until sufficient amounts
of actual data are available. Lines 9 and 10 correspond to Steps
215 through 227 of FIG. 8B for establishing a more accurate DR
strategy once sufficient data is stored.
[0126] FIG. 10 compares HVAC energy consumption models created
using the simplified approach, the neural network approach, and the
regression approach with the energy consumption simulated by the
simulation module of system 60, which is indicated as "eQuest" in
FIG. 10. For the neural network and regression approaches, the
model is obtained based on the data saved in the queues, updated
each time when new data is added into the queues, and then used for
the energy consumption estimation for the next day. For example, if
the model is updated on Tuesday, it will be used for energy
consumption estimation on Wednesday. As shown in FIG. 10, results
generated by using the simplified model are quite different from
actual simulated energy consumption. However, for neural network
and regression approaches, the estimation becomes more reliable and
accurate as more data are collected and used to learn the model.
FIG. 10(a) illustrates the modeled energy consumption using two
days of data, 10(b) illustrates modeled energy consumption using
four days of data, and 10(c) illustrates modeled energy consumption
using six days of data. As shown in these three graphs, the
difference between the models learned by using the neural network
and regression approaches is small, and the estimation by the
neural network and regression approaches becomes closer to the
actual simulated energy consumption as data from more days are
stored in the queues.
[0127] To better understand the performance, a comparison study is
made as shown by FIGS. 11 and 12 through the computational
experiment system 60, in which "None" represents a constant
thermostat setting at 72.degree., "Heuristic" stands for the
thermostat setting by using heuristic DR strategy, "Simple"
signifies the optimal thermostat setting based on the simplified
model (see Equation 5), and "Regression" represents the optimal
thermostat setting by using a substantially real-time adaptive
regression model (See Equation 8). For the heuristic algorithm, a
"nine-point" thermostat setting at 71.degree.; 72.degree.,
73.degree., 74.degree., 75.degree., 76.degree., 77.degree.,
78.degree. and 79.degree. is used, in which 71.degree. and
79.degree. are the thermostat settings corresponding to PR.sup.min
and PR.sup.max during a day, respectively. For the optimal DR
strategy algorithms, the upper and lower temperature settings are
71.degree. and 79.degree., respectively. Table 1 below shows a
comparison of the HVAC cost for five consecutive days, in which the
regression-based approach updates the HVAC energy consumption model
daily when new energy consumption data is available. FIG. 12 shows
a comparison of the HVAC cost for one month by using different
energy consumption models based on DAP tariff. As can be seen from
FIGS. 11 and 12 and Table 1, as well as other results, the
learning-based DR strategies are the most efficient among all the
DR strategies, demonstrating the effectiveness and excellent
performance of the learning based DR strategy systems.
TABLE-US-00003 TABLE 1 HVAC energy cost for five consecutive days
based on DAP Day 1 Day 2 Day 3 Day 4 Day 5 None $1.571 $1.503
$1.181 $0.867 $1.571 Heuristic $1.124 $1.028 $0.850 $0.687 $1.020
Approach Simple $1.019 $0.997 $0.795 $0.642 $0.983 Optimization
Approach Regression $1.015 $0.988 $0.789 $0.595 $0.902 Based
Approach
[0128] Table 2 below shows a comparison of the HVAC cost for a
high, mild, and low cost day for RTP and DAP tariffs. PSO and
regression-based approaches are listed separately.
TABLE-US-00004 TABLE 2 HVAC energy cost for five continuous days
Highest Mild Low RTP DAP RTP DAP RTP DAP None $5.09 $3.44 $1.60
$1.57 $1.36 $1.32 Heuristic Approach $3.77 $2.28 $1.08 $1.38 $1.04
$0.88 Optimization $3.12 $2.05 $0.93 $0.80 $0.90 $0.82 Approach
based on Eq. (5) Optimization $3.12 $1.98 $0.73 $0.76 $0.73 $0.66
Approach based on Eq. (8) PSO $3.11 $1.94 $0.77 $0.56 $0.88 $0.76
Approach based on Eq. (8)
[0129] FIG. 13 compares total energy and energy cost for the house
with and without demand response during a high DAP day on a Friday
in July. As shown in FIG. 13, with demand response, peak load is
clearly reduced and the energy usage of dryers and dishwasher is
shifted to low cost time period of the day.
[0130] Compared to conventional simplified models, the
learning-based approaches to modeling home energy consumption can
accurately capture the energy consumption of a house under
complicated weather conditions. In particular, FIGS. 14 and 15
illustrate a comparison of the HVAC energy consumption and cost
using the following DR strategy approaches: none, a heuristic
approach ("Heuristic"), a simplified optimal approach ("Model-OP")
based on Equation 5, an optimal approach based on regressed energy
consumption model ("Reg-OP"), and a PSO optimization approach based
on regressed energy consumption model ("PSO-OP"). FIG. 14
illustrates the comparison using RTP tariff for the highest RTP
tariff day in the summer, and FIG. 15 illustrates the comparison
using the DAP tariff for the highest DAP tariff in the summer. The
comparisons in FIGS. 14 and 15 show that the difference in cost
saving between the DR strategies solved by using normal nonlinear
programming technique ("Reg-OP") and the PSO method ("PSO-OP") is
small.
[0131] The flowchart and block diagrams in the figures illustrate
the architecture, functionality, and operation of possible
implementations of systems, methods and computer program products
according to various implementations of the present invention. In
this regard, each block in the flowchart or block diagrams may
represent a module, segment, or portion of code, which comprises
one or more executable instructions for implementing the specified
logical function(s). It should also be noted that, in some
alternative implementations, the functions noted in the block may
occur out of the order noted in the figures. For example, two
blocks shown in succession may, in fact, be executed substantially
concurrently, or the blocks may sometimes be executed in the
reverse order, depending upon the functionality involved. It will
also be noted that each block of the block diagrams and/or
flowchart illustration, and combinations of blocks in the block
diagrams and/or flowchart illustration, can be implemented by
special purpose hardware-based systems that perform the specified
functions or acts, or combinations of special purpose hardware and
computer instructions.
[0132] The corresponding structures, materials, acts, and
equivalents of all means or step plus function elements in the
claims below are intended to include any structure, material, or
act for performing the function in combination with other claimed
elements as specifically claimed. The description of the present
invention has been presented for purposes of illustration and
description, but is not intended to be exhaustive or limited to the
invention in the form disclosed. Many modifications and variations
will be apparent to those of ordinary skill in the art without
departing from the scope and spirit of the invention. The
implementation was chosen and described in order to best explain
the principles of the invention and the practical application, and
to enable others of ordinary skill in the art to understand the
invention for various implementations with various modifications as
are suited to the particular use contemplated.
[0133] Any combination of one or more computer readable medium(s)
may be used to implement the systems and methods described
hereinabove. The computer readable medium may be a computer
readable signal medium or a computer readable storage medium. A
computer readable storage medium may be, for example, but not
limited to, an electronic, magnetic, optical, electromagnetic,
infrared, or semiconductor system, apparatus, or device, or any
suitable combination of the foregoing. More specific examples (a
non-exhaustive list) of the computer readable storage medium would
include the following: an electrical connection having one or more
wires, a portable computer diskette, a hard disk, a random access
memory (RAM), a read-only memory (ROM), an erasable programmable
read-only memory (EPROM or Flash memory), an optical fiber, a
portable compact disc read-only memory (CD-ROM), an optical storage
device, a magnetic storage device, or any suitable combination of
the foregoing. In the context of this document, a computer readable
storage medium may be any tangible medium that can contain, or
store a program for use by or in connection with an instruction
execution system, apparatus, or device.
[0134] A computer readable signal medium may include a propagated
data signal with computer readable program code embodied therein,
for example, in baseband or as part of a carrier wave. Such a
propagated signal may take any of a variety of forms, including,
but not limited to, electro-magnetic, optical, or any suitable
combination thereof. A computer readable signal medium may be any
computer readable medium that is not a computer readable storage
medium and that can communicate, propagate, or transport a program
for use by or in connection with an instruction execution system,
apparatus, or device.
[0135] Program code embodied on a computer readable medium may be
transmitted using any appropriate medium, including but not limited
to wireless, wireline, optical fiber cable, RF, etc., or any
suitable combination of the foregoing.
[0136] Computer program code for carrying out operations for
aspects of the present invention may be written in any combination
of one or more programming languages, including an object oriented
programming language such as Java, Smalltalk, C++ or the like and
conventional procedural programming languages, such as the "C"
programming language or similar programming languages. The program
code may execute entirely on the user's computer, partly on the
user's computer, as a stand-alone software package, partly on the
user's computer and partly on a remote computer or entirely on the
remote computer or server. In the latter scenario, the remote
computer may be connected to the user's computer through any type
of network, including a local area network (LAN) or a wide area
network (WAN), or the connection may be made to an external
computer (for example, through the Internet using an Internet
Service Provider).
[0137] Aspects of the present invention are described herein with
reference to flowchart illustrations and/or block diagrams of
methods, apparatus (systems) and computer program products
according to implementations of the invention. It will be
understood that each block of the flowchart illustrations and/or
block diagrams, and combinations of blocks in the flowchart
illustrations and/or block diagrams, can be implemented by computer
program instructions. These computer program instructions may be
provided to a processor of a general purpose computer, special
purpose computer, or other programmable data processing apparatus
to produce a machine, such that the instructions, which execute via
the processor of the computer or other programmable data processing
apparatus, create means for implementing the functions/acts
specified in the flowchart and/or block diagram block or
blocks.
[0138] These computer program instructions may also be stored in a
computer readable medium that can direct a computer, other
programmable data processing apparatus, or other devices to
function in a particular manner, such that the instructions stored
in the computer readable medium produce an article of manufacture
including instructions which implement the function/act specified
in the flowchart and/or block diagram block or blocks.
[0139] The computer program instructions may also be loaded onto a
computer, other programmable data processing apparatus, or other
devices to cause a series of operational steps to be performed on
the computer, other programmable apparatus or other devices to
produce a computer implemented process such that the instructions
which execute on the computer or other programmable apparatus
provide processes for implementing the functions/acts specified in
the flowchart and/or block diagram block or blocks.
* * * * *